Neamt Scribe

A Neamt skill that converts YouTube videos into structured Markdown notes using OpenAI Whisper (local transcription) and Claude (note generation).

What it does

1. Downloads the audio track of any YouTube video via yt-dlp
2. Transcribes it locally with Whisper (no cloud transcription — runs on your machine)
3. Sends the transcript to Claude to generate structured Markdown notes with:
   • Executive summary
   • Key concepts
   • Main points
   • Important quotes
   • Data & statistics
   • Action items / takeaways
4. Saves the note to ~/.neamt/data/scribe/notes/ and shows it in the dashboard

Installation

neamt install github:kevinneamt/neamt-scribe

Or from a local directory:

neamt install ./neamt-scribe

The install script will:

• Install yt-dlp and openai-whisper via pip
• Create ~/.neamt/data/scribe/notes/

Then configure your Anthropic API key:

neamt config set anthropic_api_key sk-ant-...

Requirements

• Python 3.11+
• ffmpeg (recommended for best audio quality — brew install ffmpeg)
• An Anthropic API key

Permissions requested

Permission	Why
internet	yt-dlp needs to download audio from YouTube
filesystem:write	Saves .md notes to ~/.neamt/data/scribe/notes/
anthropic_api	Calls Claude to generate structured notes from the transcript

Usage

Dashboard

Open the Neamt dashboard (neamt start) and click Scribe in the sidebar.

Paste a YouTube URL → click Process → wait (transcription takes ~1× real-time on CPU) → the note appears in the list and opens automatically.

Programmatic (via bridge)

from neamt.loader import discover_skills
from neamt.bridge import handle_call

skills = {s.manifest.id: s for s in discover_skills()}
scribe = skills["scribe"]

# Process a video
result = handle_call(scribe, "process_video", {"url": "https://youtube.com/watch?v=..."})
# → {"status": "ok", "result": {"status": "done", "title": "...", "filename": "...", "word_count": N}}

# List notes
result = handle_call(scribe, "list_notes", {})

# Read a note
result = handle_call(scribe, "get_note", {"filename": "2026-06-15_some-title.md"})

# Delete a note
result = handle_call(scribe, "delete_note", {"filename": "2026-06-15_some-title.md"})

Notes storage

~/.neamt/data/scribe/
├── notes/
│   ├── 2026-06-15_video-title.md
│   └── ...
└── .tmp/        ← audio files during processing (auto-deleted)

Uninstall

neamt uninstall scribe

You will be asked whether to delete your notes before the skill files are removed.

Model

Notes are generated with claude-haiku-4-5-20251001 by default (fast and cheap). Whisper uses the base model by default (~150 MB, downloads on first use).